The AI Winner Probably Won't Ask You To Download Anything

The ChatGPT Moment

We’ve all seen the mind-blowing stats of ChatGPT adoption. At 3 years old, ChatGPT already had over 800 million weekly users (Business Insider, 2025) - a meteoric rise in success that created history in terms of the velocity of app adoption.

But here’s the thing… they built something amazing and groundbreaking that brought “all the boys to the yard” at a time when no one had seen such an interesting product. Years later, it’s not that unique of a technology and competes with other products like Claude, Grok, Gemini, and other open-weight options (Artificial Analysis, 2026).

Models Must Not Suck

I’m not saying that all models just require good distribution - looking at you Siri. Apple delayed their AI-enhanced Siri rollout (Axios, 2025) because they know there is a caveat to the “distribution always wins” thesis: a shitty model distributed at scale does more harm than good.

There is a clear benchmark of “good enough” that a model needs to meet to tip the edge into distribution. But a lot of those models are there already. Even open-weight models like GLM-5.2 have shown that there is simply no moat to achieving “good enough” status for these models (Artificial Analysis, 2026).

Buy a nice enough computer, lots of graphics cards, and a good power management system and poof! - you can now be running inference in your home. Based on my experience, you are more likely to have your wife veto the expense than you are to fail at setting up a decent, locally run system that can perform on par with GPT 5.5 for coding tasks.

Bring The Tools Into My Ecosystem

I find that most of the work I do with AI models is NOT better prompting - it’s usually just making sure the model has all the correct data from my systems and access to do the things I need it to do.

All the time invested into building skills or CLI tools for my agents is really just solving this problem: the model needs to have my data, context, knowledge AND be able to act within my desired environments.

Look - I get your new tool is great. I just don’t want to take all the time to research and buy a new tool, connect or upload my data, or operate in a new ecosystem. There is real inertia to staying in my existing software environments (please stop using that roadrunner email address).

Additionally, most of my data is already held within someone else’s ecosystem (Google has all my email, calendars, working files from over a decade - Apple has my contacts, text message history, app behavior).

The incumbents know this when talking about their AI products: Google presents Gemini across Gmail, Docs, Sheets, Drive, Meet, and other Workspace surfaces (Google Workspace AI), while Apple describes Apple Intelligence around personal context, app actions, and on-screen awareness (Apple Developer).

Sometimes convenience wins. This is noticeable if you just look at Gemini’s adoption stats relative to ChatGPT recently: Sensor Tower data reported by TechRadar put Gemini at 27.7% AI assistant share vs ChatGPT’s 46.4% in May 2026, with Google’s ecosystem integrations called out as an advantage (TechRadar, 2026).

What Next?

The awkward position now belongs to the model-first companies like OpenAI and Anthropic - who need to master distribution and monetization to keep pace with their expenditures.

OpenAI and Anthropic may have the best engines, but Google, Apple, Microsoft, and Meta own the roads. Do the labs get acquired? Do they become captive providers inside someone else’s ecosystem? Do they build enough distribution of their own before the incumbents catch up?

I don’t know.

But my bet is simpler: the winner won’t require my wife to install a single thing on her phone or computer to experience the benefits. It will just show up in the tools she already uses.